CN104200444A - Image restoring method based on symmetric sample pieces - Google Patents

Abstract

The invention provides an image restoring method based on symmetric sample pieces. The image restoring method comprises the following steps: original damaged image pre-processing, damaged zone segmentation and damaged zone restoration, wherein the original damaged image pre-processing comprises image Gauss smoothening, image graying processing; then damaged zone segmentation and damaged zone restoration are conducted; the damaged zone restoration is a recycling process which comprises the following steps: acquiring boundary points in the damaged zone, acquiring a point with highest priority in the boundary of the damaged zone and the first restored damaged piece as well as selecting a method for searching optimal symmetric sample pieces; the method for searching the optimal symmetric sample pieces comprises two methods, namely, searching symmetric sample pieces at any direction or searching symmetric sample pieces in eight directions, and each of the two methods comprises the following steps: updating pixel of the damaged pieces, updating area of the damaged zone and judging whether the area of the damaged zone is zero, if the area of the damaged zone is zero, stopping the restoration; if the area of the damaged zone is not zero, returning to the acquired boundary points for conducting the recycling process. The restoring experiment shows the effectiveness of the image restoring method.

Description

Image repair method based on balanced sample piece

Technical field

The invention belongs to Digital Image Processing and digital picture and repair field, be specifically related to the image repair method based on balanced sample piece.

Background technology

Since " image repair " concept in 2000, introduce after Digital Image Processing, short more than ten years, digital picture reparation is widely used, and mainly comprises: ancient painting numeral is restored, recovered word in damaged photo and film, image and object and removes etc.So-called digital picture reparation is exactly the process of information filling that defect area on image is carried out, and its object is exactly for to there being damaged image to recover, and visually cannot perceive the once damaged or effect that has been repaired of image.In image repair technology, a class is reparation algorithm based on partial differential equation or the variation recovery technique based on several picture model, utilizes the marginal information for the treatment of repairing area, repairs small scale deletion problem.Another kind of is image completion technology for the bulk drop-out of blank map picture, comprise two kinds of methods: a kind of is method based on decomposing, by picture breakdown, be structure and texture part, wherein structure division is repaired with image repair algorithm, and texture part is filled by the synthetic method of texture.Also having a kind of is that block-based Future Opportunities of Texture Synthesis is filled the information of losing, it is from zone boundary to be repaired, to choose a pixel, and centered by this point, according to the textural characteristics of image, choose sizeable texture block, then in the surrounding of multiblock to be repaired, find the most similar with it Texture Matching piece and replace.Yet image repair remains a difficult thing.

Existing Digital Image Inpainting can't effectively be repaired a large amount of symmetric graph pictures, as Tangka's image, mural painting etc., a lot of images have certain symmetrical structure, in image, some crucial pieces is damaged, as one side ear of Sakyamuni Mo Nifo in a width Tangka loses, if only utilize ear's peripheral information reparation in this image, cannot obtain needed result.

Summary of the invention

In order to utilize, be repaired the symmetric information possessing in image, repair damaged area, the invention provides a kind of image repair method based on balanced sample piece, it is a kind of effective digitized image restorative procedure that utilizes the information filling damaged area of damaged area a certain symmetry direction on image.

In order to realize the object of the invention, the technical scheme that the present invention takes is: a kind of image repair method based on balanced sample piece, comprise to former breakage image carry out pre-service, damaged area is cut apart and repair damaged area, concrete steps are as follows:

A. former breakage image is carried out to pre-service

A.1 image is carried out to Gaussian smoothing

Because Gaussian smoothing edge keeps better, conventionally using following 5 * 5 or 3 * 3 Gaussian smoothing template mask image:

$\begin{array}{cc}\frac{1}{273}\×\left[\begin{array}{ccccc}1& 4& 7& 4& 1\\ 4& 16& 26& 16& 4\\ 7& 26& 41& 26& 7\\ 4& 16& 26& 16& 4\\ 1& 4& 7& 4& 1\end{array}\right]& \frac{1}{16}\×\left[\begin{array}{ccc}1& 2& 1\\ 2& 4& 2\\ 1& 2& 1\end{array}\right]\end{array}$

A.2 image gray processing is processed

Adopt mean value method by coloured image gray processing, obtain R, the G of each pixel of image, the mean value of tri-component pixel values of B, and generate its gray level image;

B. damaged area is cut apart

Conventional image partition method comprises thresholding method, boundary segmentation method and region growing method Equal method; The present invention adopts the dividing method of region growing, and its process is:

B.1. adopt the method for man-machine interaction, in being repaired the damage zone of image, choose arbitrarily a bit as Seed Points coordinate (x _r, y _r), and on gray level image, record its pixel value p _r(x _r, y _r);

B.2. create a two-value template image mask onesize with former figure, by pixel value p in template image mask _r(x _r, y _r) being made as 1, rest of pixels is set to 0;

B.3. on gray level image, pass through algorithm 1, draw the some set that in template image mask, pixel value is 1, with regard to the damaged area of corresponding original image, be namely partitioned into the damaged area of original image;

The step that algorithm 1 comprises:

(1) create a storehouse;

(2) obtain Seed Points (x _r, y _r) and pixel value p _r(x _r, y _r);

(3) with Seed Points (x _r, y _r) centered by, calculate its pixel value p _r(x _r, y _r) and eight neighborhood territory pixel value p _i(x _i, y _i) poor, if | p _r(x _r, y _r)-p _i(x _i, y _i) | < M, (i=1,2,3,4,5,6,7,8), the value of M is determined according to experiment, generally get 10, and on mask opposite position, pixel value is at 0 o'clock, by point (x _i, y _i) being pressed into storehouse, the pixel value on mask relevant position is made as 1;

(4) judge that whether storehouse is empty, if not for sky, take out a pixel from storehouse, is used as it as point (x _r, y _r), go back to step (3), otherwise turn to step (5);

(5) finish;

C. repair damaged area

C.1 obtain the frontier point of damaged area

C.1.1 create a queue, be used for storing damaged area frontier point;

C.1.2 obtain damaged area frontier point: to two-value template image mask, the Wei Yuantu damaged area, region that wherein pixel value is 1, traversal view picture mask image, if the pixel value of pixel is 1, and its eight adjacent pixel values has at least one to be 0, the pixel that is just 1 by pixel value deposits in queue, and the queue finally obtaining is exactly the frontier point of damaged area;

C.2 obtain the point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first

Figure 2 shows that right of priority peak on border and the damaged piece of repairing at first calculate schematic diagram, Ι represents full figure, and Ω represents damaged area, and δ Ω represents the border of damaged area Ω, the unbroken part of Φ presentation video, and p represents the point on the δ Ω of border, damaged area, n _pa unit normal vector at p place, represent the isophot curve direction that p is ordered, Ψ _pbe on border centered by the p point, comprised the damaged piece of putting in point in Ω and Φ; The point that on border, damaged area, right of priority is the highest and the damaged piece computation process of the reparation at first centered by this point are as follows:

C.2.1 on computation bound, put the right of priority of p, by formula (1), calculated,

P(p)＝C(p)D(p) (1)

In formula (1): C (p) represents the confidence factor that p is ordered, D (p) represents the data factor that p is ordered, and C (p) is calculated by formula (2),

$C\left(p\right)=\frac{{\mathrm{\&Sigma;}}_{q\&Element;{\mathrm{\&Psi;}}_p\&cap;(I-\mathrm{\&Omega;})}C\left(q\right)}{\left|{\mathrm{\&Psi;}}_p\right|}---\left(2\right)$

In formula (2): | Ψ _p| be Ψ _parea, in image confidence factor a little according to formula (3) initialization,

$C\left(k\right)=\left\{\begin{array}{cc}0& \&ForAll;k\&Element;\mathrm{\&Omega;}\\ 1& \&ForAll;k\&Element;I-\mathrm{\&Omega;}\end{array}\right.---\left(3\right)$

In formula (1), D (p) through type (4) calculates,

$D\left(p\right)=\frac{|{\&dtri;I}_p^{\&perp;}\&CenterDot;n_p|}{\mathrm{\&alpha;}}---\left(4\right)$

Wherein α is normalized factor, is the maximal value of image gray levels;

C.2.2 on computation bound right of priority a little, thereby and the size of fiducial value obtain the highest point of right of priority, be designated as p ₀;

C.2.3 obtain with p ₀centered by the piece of repairing at first this piece size according to damaged area texture structure around, select and sizable of texture structure, adopt the mode of man-machine interaction, the length of side of piece can be the odd number between 3 to 99, the piece of 3 * 3 to 99 * 99 sizes, generally selects 9 * 9,11 * 11 to 33 * 33; Then enter step c 3; If image have left and right, upper and lower, upper right with all directions of lower-left, upper left and bottom right to one of symmetrical situation, select based on all directions to balanced sample piece restorative procedure, otherwise be chosen in the restorative procedure that any direction is searched for balanced sample piece;

In fact, the method for any direction search balanced sample piece reparation has ubiquity, can replace the reparation to balanced sample piece based on all directions completely; If but piece image is symmetrical, there is damaged area in left side, based on faster to the reparation speed of balanced sample piece from all directions, it is relatively consuming time that any direction is searched for the method that best balanced sample piece repairs; So enter after step c 3, can be chosen in any direction search balanced sample piece or based on from all directions to the image repair mode of balanced sample piece, once a kind of selected method is repaired according to selected this method circulation until repair and finish;

C.3 search in any direction best balanced sample piece or in the image repair of the best balanced sample pieces of eight direction findings

C.3.1 find in any direction best balanced sample piece

C.3.1.1 in I-Ω with arbitrfary point p ₁centered by sample block choose

As shown in Figure 3, p ₀the central point of repairing at first piece, p ₁be the central point of sample block to be found, obtain p ₁and p ₀line and horizontal direction angle theta, utilize formula (5) to obtain p ₁rotate to and p ₀p on same level line position ₂coordinate:

$\left\{\begin{array}{c}{x}_2=(x_1-x_0)\cos \mathrm{\&theta;}+(y_1-y_0)\sin \mathrm{\&theta;}+x_0\\ y_2=-(x_1-x_0)\sin \mathrm{\&theta;}+(y_1-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(5\right)$

In formula (5): x ₀, x ₁and x ₂respectively p ₀, p ₁and p ₂horizontal ordinate, y ₀, y ₁and y ₂respectively p ₀, p ₁and p ₂ordinate;

Acquisition is with p ₂centered by, size and identical and with become the sample block of horizontal symmetrical take 3 * 3 as example, the corresponding relation of damaged piece and sample block is as shown in Fig. 4 and Fig. 5, and Fig. 4 is damaged piece fig. 5 represents horizontal symmetrical sample block two piece symmetric positions of numeral wherein; Recycling formula (6) obtains sample block the position at opposite spin θ angle:

$\left\{\begin{array}{c}x\text{'}=(x-x_0)\cos \mathrm{\&theta;}-(y-y_0)\sin \mathrm{\&theta;}+x_0\\ y\text{'}=(x-x_0)\sin \mathrm{\&theta;}+(y-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(6\right)$

In formula (6): before (x, y) represents reverse rotation the coordinate of middle arbitrfary point, (x ', y ') represent behind rotation θ angle the coordinate of middle corresponding point; after rotation, with point corresponding relation and with corresponding relation identical, obtain with p ₁centered by balanced sample piece;

C.3.1.2 at I-Ω, search for best balanced sample piece, by formula (7), calculated,

$\mathrm{\&Psi;}\hat{p}=\arg \underset{p_1\&Element;I-\mathrm{\&Omega;}}{\min }d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)---\left(7\right)$

In formula (7): represent best balanced sample piece, with p ₁centered by sample block, represent multiblock to be repaired at first with balanced sample piece between similarity measurement, by formula (8), calculate,

$d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2}---\left(8\right)$

In formula (8): x _ijrepresent the pixel value of mid point, y _ijrepresent the pixel value of mid point, the size of piece determines the value of m, if select the size of piece, is 3 * 3, the value of m is 3;

C.3.1.3 the renewal of damaged piece

C.3.1.3.1 according to the corresponding relation in Fig. 4 and Fig. 5, by damaged piece in best balanced sample piece for damaged pixel value in respective pixel value replace;

C.3.1.3.2 upgrade the confidence factor of previous step filler pixels point, by formula (9), calculated,

$\begin{array}{cc}C\left(p\right)=C\left(\hat{p}\right)& \&ForAll;p\&Element;\mathrm{\&Psi;}\hat{p}\&cap;\mathrm{\&Omega;}\end{array}---\left(9\right)$

The pixel value of corresponding point in mask on former breakage image filling point position is updated to 0 simultaneously;

C.3.1.4 upgrade the area of damaged area, calculate the pixel count in damaged area after filling

Because calculating the damaged piece that right of priority is the highest is centered by frontier point at every turn, so existing unbroken image slices vegetarian refreshments has again the point in damaged area in damaged piece, and the point in these damaged area is filled by the optimal sample piece corresponding point of finding on symmetry direction in this breakage piece, along with filling the area of damaged area, dwindle gradually; Therefore, every filling once will be recalculated the area of damaged area afterwards, and statistics is filled the number of pixels of rear damaged area, to judge whether repair process finishes; If damaged area area is zero, reparation completes, otherwise repeats to do step c 1, step c 2 and step c 3.1;

C.3.2 at the best balanced sample piece of eight direction findings

As shown in Figure 6, in figure, the left ear of the figure of buddha is damaged, right ear is intact, can utilize symmetrical property to repair fast left ear; Eight arrows represent the damaged piece Ψ centered by p _peight directions, from these eight direction finding balanced sample pieces, each direction is found out a similar sample block of symmetry, then from wherein finding out best balanced sample piece, Ψ _qΨ _pbest balanced sample piece; Calculation procedure is as follows:

C.3.2.1 at eight similar balanced sample pieces of direction finding

From eight directions, find symmetrical similar sample block by formula (10), calculated:

$\begin{array}{cc}\mathrm{\&Psi;}{\hat{q}}_i=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(10\right)$

In formula (10): represent Ψ _pwith similarity measure, subscript 1,2,3,4,5,6,7 and 8 represent respectively a left side, upper left, upper, upper right, the right side, bottom right, under, lower left; 1, the Comparability of 2,3 and 4 four directions formula (11) calculating for amount, the Comparability of 5,6,7 and 8 four directions is formula (12) calculating for amount:

$\left\{\begin{array}{c}{d}_1(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_1=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_2(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_2)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_3(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_3)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_4(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_4)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(11\right)$

$\left\{\begin{array}{c}{d}_5(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_5=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_6(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_6)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_7(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_7)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_8(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_8)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(12\right)$

In formula (11), (12): x _ijrepresent Ψ _pthe pixel value of mid point, y _ijrepresent the pixel value of mid point;

C.3.2.2 the calculating of best balanced sample piece

From eight similar sample block of symmetry, find Best similarity balanced sample piece, can calculate according to formula (13):

$\begin{array}{cc}\mathrm{\&Psi;q}=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{\hat{q}}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(13\right)$

C.3.2.3 upgrade the damaged pixel of damaged piece

C.3.2.3.1 according to following rule Ψ _qin pixel value upgrade Ψ _pmiddle damaged pixel, when pixel is at Ψ _pin scanning sequency be from left to right, from top to bottom time, corresponding Ψ _qin scanning sequency be (referring to table 1): when the direction at best balanced sample piece place is 1 or 5, pixel scanning sequency is from right to left, from top to bottom in best balanced sample piece; When the direction at best balanced sample piece place is 2 or 6, in best balanced sample piece, pixel scanning sequency is from top to bottom, from right to left; When the direction at best balanced sample piece place is 3 or 7, in best balanced sample piece, pixel scanning sequency is from left to right, from top to bottom; When the direction at best balanced sample piece place is 4 or 8, in best balanced sample piece, pixel scanning sequency is from top to bottom, from left to right; And by Ψ _pmiddle damaged pixel Ψ _qthe pixel value of middle correspondence is filled;

The best balanced sample piece pixel scanning sequency table of eight directions of table 1.

Best balanced sample piece place direction	Pixel scanning sequency in best balanced sample piece
		1 or 5	From right to left, from top to bottom
2 or 6	From top to bottom, from right to left
		3 or 7	From left to right, from top to bottom
4 or 8	From top to bottom, from left to right

Take block size 3 * 3 as example, optimal sample piece left to time, Fig. 7 has described the process that damaged piece pixel is upgraded, wherein a represents damaged piece Ψ _p, b represents best balanced sample piece Ψ _q, the numeral scanning sequency in a, b, the corresponding pixel of digital identical expression, if Ψ _pin 4 belong to damaged pixel, just use Ψ _qthe pixel value of middle position 4 is filled;

C.3.2.3.2 upgrade the confidence factor of previous step filler pixels point, by formula (9), calculated; The pixel value of corresponding point in mask on image filling point position is updated to 0 simultaneously;

C.3.2.4 upgrade the area of damaged area, if damaged area area is zero, reparation completes, otherwise repeats to do step c 1, step c 2 and step c 3.2.

The present invention is for a large amount of breakage image reparations, and reparative experiment has illustrated the validity of the inventive method.

Technique effect of the present invention is as follows: the present invention has good repairing effect to having certain symmetric breakage image, the prerequisite of repairing is to be accurately partitioned into damaged area, if damaged area segmentation result is not good enough, natural repairing effect is bad, and aftermentioned embodiment part example is by pre-service, be accurately partitioned into the image of damaged area.

Accompanying drawing explanation

Fig. 1 is the image repair process flow diagram based on balanced sample piece,

Fig. 2 is that on border, right of priority peak and the damaged piece repaired at first calculate schematic diagram,

Fig. 3 is the balanced sample piece conversion schematic diagram of any direction,

Fig. 4 is 3 * 3 sample block schematic diagram,

Fig. 5 is 3 * 3 damaged piece schematic diagram,

Fig. 6 is eight direction exemplary plot of damaged piece,

Fig. 7 upgrades damaged piece pixel schematic diagram with 3 * 3 balanced sample pieces,

Fig. 8 is symmetrical breakage image visually,

Fig. 9 is to the size of balanced sample piece, piece, to be from all directions the reparation result of 33 * 33 o'clock Fig. 8,

Figure 10 is Manjusri's Bodhisattva Tangka image, and head slightly tilts, pileum part is damaged,

Figure 11 is the reparation result of any direction balanced sample piece to Figure 10,

Figure 12 is the reparation result to Figure 10 to balanced sample piece from all directions.

Embodiment

Thereby the present invention mainly solves breakage image, there is certain symmetry and carry out the reparation problem based on balanced sample piece; Technical scheme of the present invention is the image repair method based on balanced sample piece, and whole process comprises repairs pre-service, the cutting apart with damaged area of damaged area to former breakage image.Referring to Fig. 1: former breakage image is carried out to pre-service and comprise: to image carry out Gaussian smoothing, image gray processing is processed; Then entering damaged area cuts apart with damaged area and repairs; Repair damaged area is the process of an iterative cycles, this process comprise obtains damaged area frontier point, obtain the point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first, and select to search for best balanced sample block method; Select the best balanced sample block method of search to comprise that any direction searches for best balanced sample piece or two kinds of methods of the best balanced sample piece of eight direction findings, these two kinds of methods all comprise to be upgraded damaged pixel, the area of renewal damaged area of damaged piece and judges whether the area of damaged area is zero, to repair and finish; No, get back to the frontier point repetitive cycling that obtains damaged area.

Embodiment 1

1. the former breakage image shown in couple Fig. 8 carries out pre-service

Fig. 8 is Sakyamuni Buddha Tangka coloured image, and left side ear, face, eyes and part eyebrow are damaged; Former breakage image is carried out to pre-service and mainly comprise image is carried out to level and smooth and gray processing processing, smoothing processing is in order to eliminate noise, so that the better effects if that image is cut apart; Because cutting apart on gray level image of damaged area carried out, so preprocessing part has comprised, the gray processing of former breakage image to be processed, preprocessing process is: 1.1 pairs of images carry out Gaussian smoothing

Because Gaussian smoothing edge keeps better, the present embodiment is selected 5 * 5 Gaussian smoothing template mask images:

$\frac{1}{273}\&times;\left[\begin{array}{ccccc}1& 4& 7& 4& 1\\ 4& 16& 26& 16& 4\\ 7& 26& 41& 26& 7\\ 4& 16& 26& 16& 4\\ 1& 4& 7& 4& 1\end{array}\right]$

1.2 image gray processings are processed

Adopt mean value method by coloured image gray processing, obtain R, the G of each pixel of image, the mean value of tri-component pixel values of B, and generate the gray level image of original color image.

2. damaged area is cut apart

Conventional image partition method comprises thresholding method, boundary segmentation method and region growing method Equal method; In the present invention, damaged area cut apart the dividing method that adopts region growing, its process is:

2.1. adopt the method for man-machine interaction, in being repaired the damage zone of image, choose arbitrarily a bit as Seed Points coordinate (x _r, y _r), and record its pixel value p _r(x _r, y _r);

2.2. create a two-value template image mask onesize with former figure, by p in template image mask _r(x _r, y _r) being made as 1, rest of pixels is set to 0;

2.3. by algorithm 1, obtain the some set that in template image mask, pixel value is 1, the damaged area of the corresponding original images of these pixel set, has been partitioned into the damaged area of original image.

The concrete steps of algorithm 1 are:

(1) create a storehouse;

(2) obtain Seed Points (x _r, y _r) and pixel value p _r(x _r, y _r);

(3) with Seed Points (x _r, y _r) centered by, calculate its pixel value p _r(x _r, y _r) and eight neighborhood territory pixel value p _r(x _i, y _i) poor, if | p _r(x _r, y _r)-p _i(x _i, y _i) | < M, (i=1,2,3,4,5,6,7,8), the value of M is determined according to experiment, be generally taken as 10, and on template image mask relevant position, pixel value is at 0 o'clock, by point (x _i, y _i) being pressed into storehouse, the pixel value on mask relevant position is made as 1;

(4) judge that whether storehouse is empty, if not for sky, take out a pixel from storehouse, is used as it as point (x _r, y _r), forward (3) step to, otherwise turn to (5);

(5) finish.

3. repair damaged area

Repair damaged area based on balanced sample piece is the process of a circulation, comprise obtain damaged area frontier point, obtain point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first, the best balanced sample pieces of eight direction findings, upgrade the damaged pixel of damaged piece and upgrade the area of damaged area, until the area of damaged area is when being zero, whole repair process finishes, thereby obtains the result images after repairing; That is to say, on gray level image, be partitioned into behind damaged area, on two-value template image mask, represent damaged area, and on original image, repair damaged area; Along with each filling of damaged area is repaired, two-value template image mask also constantly updates the upper damaged area representing, until mask completes reparation when above the area of represented damaged area is zero.Process following steps are carried out:

3.1 obtain the frontier point of damaged area

3.1.1 create a queue, be used for storing damaged area frontier point;

When 3.1.2 image is cut apart, obtain bianry image mask, the Wei Yuantu damaged area, region that wherein pixel value is 1; Traversal view picture mask image, if the pixel value of pixel is 1, and its eight adjacent pixel values has at least one to be 0, and the pixel that is just 1 by pixel value deposits in queue, and the queue finally obtaining is exactly the frontier point set of damaged area.

3.2 obtain the point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first

Figure 2 shows that right of priority peak on border and the damaged piece of repairing at first calculate schematic diagram, Ι represents full figure, and Ω represents damaged area, δ Ω represents the border of damaged area Ω, the unbroken part of Φ presentation video, Ρ represents the point on the δ Ω of border, damaged area, n _pa unit normal vector at p place, represent the isophot curve direction that p is ordered, Ψ _pbe on border centered by the p point, comprised the damaged piece of putting in point in Ω and Φ; The point that on border, damaged area, right of priority is the highest and the damaged piece computation process of the reparation at first centered by this point are as follows:

3.2.1 on computation bound, put the right of priority of p, by (1) formula, obtained:

P(p)＝C(p)D(p) (1)

In formula (1), C (p) represents the confidence factor that p is ordered, and D (p) represents the data factor that p is ordered, and C (p) is determined by formula (2):

$C\left(p\right)=\frac{{\mathrm{\&Sigma;}}_{q\&Element;{\mathrm{\&Psi;}}_p\&cap;(I-\mathrm{\&Omega;})}C\left(q\right)}{\left|{\mathrm{\&Psi;}}_p\right|}---\left(2\right)$

Wherein | Ψ _p| be Ψ _parea, i.e. piece Ψ _pthe quantity of interior pixel, in image confidence factor a little according to formula (3) initialization:

$C\left(k\right)=\left\{\begin{array}{cc}0& \&ForAll;k\&Element;\mathrm{\&Omega;}\\ 1& \&ForAll;k\&Element;I-\mathrm{\&Omega;}\end{array}\right.---\left(3\right)$

In formula (1), D (p) calculates by formula (4):

$D\left(p\right)=\frac{|{\&dtri;I}_p^{\&perp;}\&CenterDot;n_p|}{\mathrm{\&alpha;}}---\left(4\right)$

Wherein minute two vectorial inner products of subrepresentation of (4) formula take absolute value again, and α is normalized factor, is the maximal value of image gray levels.

3.2.2 on computation bound right of priority a little, thereby and the size of fiducial value obtain the highest point of right of priority, be designated as p;

3.2.3 obtain the piece Ψ repairing at first centered by p _p, this piece Ψ _psize according to damaged area texture structure around, select and sizable of texture structure, adopt the mode of man-machine interaction, the length of side of piece can be the odd number between 3 to 99, i.e. the piece of 3 * 3 to 99 * 99 sizes, it is 33 that the present embodiment is selected the length of side of piece; Then enter step 3.3; Because the image of Fig. 8 has the situation of left and right directions symmetry, thus select based on from all directions to balanced sample piece repair.

3.3 image repair at the best balanced sample piece of eight direction findings

3.3.1 at the best balanced sample piece of eight direction findings

As shown in Figure 8, in figure, the left side ear of the figure of buddha, face, eyes and part eyebrow are damaged; Find the highest multiblock Ψ to be repaired of right of priority _pall directions to best balanced sample piece Ψ _qstep as follows:

3.3.1.1 at eight similar balanced sample pieces of direction finding

The similar sample block of symmetry that finds Ψ p from eight directions by formula (10), calculated:

$\begin{array}{cc}\mathrm{\&Psi;}{\hat{q}}_i=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(10\right)$

In formula (10): represent Ψ _pwith similarity measure, subscript 1,2,3,4,5,6,7 and 8 represent respectively a left side, upper left, upper, upper right, the right side, bottom right, under, lower left, 1, the Comparability of 2,3 and 4 four directions formula (11) calculating for amount, the Comparability of 5,6,7 and 8 four directions is formula (12) calculating for amount:

$\left\{\begin{array}{c}{d}_1(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_1=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_2(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_2)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_3(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_3)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_4(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_4)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(11\right)$

$\left\{\begin{array}{c}{d}_5(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_5=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_6(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_6)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_7(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_7)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_8(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_8)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(12\right)$

In formula (11), (12): x _ijrepresent Ψ _pthe pixel value of mid point, y _ijrepresent the pixel value of mid point;

3.3.1.2 the calculating of best balanced sample piece

From eight similar sample block of symmetry, find Best similarity balanced sample piece, can calculate according to formula (13):

$\begin{array}{cc}\mathrm{\&Psi;q}=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{\hat{q}}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(13\right)$

3.3.1.3 upgrade the damaged pixel of damaged piece

3.3.1.3.1 according to following rule Ψ _qin pixel value upgrade Ψ _pmiddle damaged pixel, when pixel is at Ψ _pin scanning sequency be from left to right, from top to bottom time, corresponding Ψ _qin scanning sequency be (referring to table 1): when the direction at best balanced sample piece place is 1 or 5, pixel scanning sequency is from right to left, from top to bottom in best balanced sample piece; When the direction at best balanced sample piece place is 2 or 6, in best balanced sample piece, pixel scanning sequency is from top to bottom, from right to left; When the direction at best balanced sample piece place is 3 or 7, in best balanced sample piece, pixel scanning sequency is from left to right, from top to bottom; When the direction at best balanced sample piece place is 4 or 8, in best balanced sample piece, pixel scanning sequency is from top to bottom, from left to right; And by Ψ _pmiddle damaged pixel Ψ _qthe pixel value of middle correspondence is filled;

3.3.1.3.2 upgrade the confidence factor of previous step filler pixels point, by formula (9), calculated:

$\begin{array}{cc}C\left(p\right)=C\left(\hat{p}\right)& \&ForAll;p\&Element;\mathrm{\&Psi;}\hat{p}\&cap;\mathrm{\&Omega;}\end{array}---\left(9\right)$

3.3.1.4 upgrade the area of damaged area, if damaged area area is zero, reparation completes, otherwise repeats to do step 3.1, step 3.2 and step 3.3.

The present embodiment is to breakage image shown in Fig. 8, adopt all directions to the method reparation of balanced sample piece, the size of selection piece is 33 * 33, result as shown in Figure 9, to this breakage image, if select the size of piece between 33 * 33 and 43 * 43, vestiges of almost not seeing reparation as shown in Figure 9 all; If select piece to be less than 33 * 33, to be greater than 43 * 43, reparation result is all undesirable.Illustrate relevant with the selection of sample block size to the repairing effect of balanced sample piece from all directions.

Embodiment 2, Figure 10 shows that Manjusri's Bodhisattva Tangka image, and a side of pileum has breakage, and head has a bit, and the former breakage image shown in Figure 10 is carried out to pre-service to the method providing according to embodiment 1 step 1 and step 2 and damaged area is cut apart; Then enter step:

3. repair damaged area

3.1 obtain the frontier point of damaged area, with embodiment 1;

3.2 obtain the point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first, with embodiment 1;

3.3 find best balanced sample piece in any direction carries out image repair

3.3.1 find in any direction best balanced sample piece

3.3.1.1 in I-Ω with arbitrfary point p ₁centered by sample block choose

As shown in Figure 3, p ₀the central point of repairing at first piece, p ₁be the central point of sample block to be found, obtain p ₁and p ₀line and horizontal direction angle theta, utilize formula (5) to obtain p ₁rotate to and p ₀p on same level line position ₂coordinate:

$\left\{\begin{array}{c}{x}_2=(x_1-x_0)\cos \mathrm{\&theta;}+(y_1-y_0)\sin \mathrm{\&theta;}+x_0\\ y_2=-(x_1-x_0)\sin \mathrm{\&theta;}+(y_1-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(5\right)$

In formula (5): x ₀, x ₁and x ₂respectively p ₀, p ₁and p ₂horizontal ordinate, y ₀, y ₁and y ₂respectively p ₀, p ₁and p ₂ordinate;

Acquisition is with p ₂centered by, size and identical and with become the sample block of horizontal symmetrical recycling formula (6) obtains sample block the position at opposite spin θ angle:

$\left\{\begin{array}{c}x\text{'}=(x-x_0)\cos \mathrm{\&theta;}-(y-y_0)\sin \mathrm{\&theta;}+x_0\\ y\text{'}=(x-x_0)\sin \mathrm{\&theta;}+(y-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(6\right)$

In formula (6): before (x, y) represents reverse rotation the coordinate of middle arbitrfary point, (x ', y ') represent behind rotation θ angle the coordinate of middle corresponding point; after rotation, with point corresponding relation and with corresponding relation identical, obtain with p ₁centered by balanced sample piece;

3.3.1.2 at I-Ω, search for best balanced sample piece, by formula (7), calculated,

$\mathrm{\&Psi;}\hat{p}=\arg \underset{p_1\&Element;I-\mathrm{\&Omega;}}{\min }d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)---\left(7\right)$

In formula (7): represent best balanced sample piece, the sample block centered by p1, represent multiblock to be repaired at first with balanced sample piece between similarity measurement, by formula (8), calculate,

$d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2}---\left(8\right)$

In formula (8): x _ijrepresent the pixel value of mid point, yij represents the pixel value of mid point, the size of piece is 33 * 33;

3.3.1.3 the renewal of damaged piece

3.3.1.3.1 by damaged piece in best balanced sample piece for damaged pixel value in respective pixel value replace;

3.3.1.3.2 upgrade the confidence factor of previous step filler pixels point, can be calculated by formula (9),

$\begin{array}{cc}C\left(p\right)=C\left(\hat{p}\right)& \&ForAll;p\&Element;\mathrm{\&Psi;}\hat{p}\&cap;\mathrm{\&Omega;}\end{array}---\left(9\right)$

The pixel value of corresponding point in mask on former breakage image filling point position is updated to 0 simultaneously;

3.3.1.4 upgrade the area of damaged area, calculate the pixel count in damaged area after filling

Statistics is filled the number of pixels of rear damaged area, to judge whether repair process finishes; If damaged area area is zero, reparation completes, otherwise repeats to do step 3.1, step 3.2 and step 3.3.

Adopt the present embodiment to the result of Figure 10 reparation as shown in figure 11, can find out that repairing effect is fairly good.

Figure 12 is for adopting the reparation result to Figure 10 to balanced sample piece restorative procedure from all directions, effect is flaw well but slightly also, illustrate the breakage image shown in Figure 10 do not have from all directions to one of symmetry, so select from all directions to best balanced sample piece restorative procedure effect not as the best balanced sample piece of any direction restorative procedure.

In addition, breakage image shown in Fig. 8 is also adopted to the method reparation of the best balanced sample piece of any direction, the size of piece is between 33 * 33 and 43 * 43 time, repair result and also almost do not see the vestige of reparation, just adopt the restorative procedure of the best balanced sample piece of any direction than consuming time to best balanced sample piece restorative procedure from all directions; If select other big or small sample block, reparation result is also all undesirable.

Claims (2)

1. the image repair method based on balanced sample piece, it is characterized in that comprising to former breakage image carry out pre-service, damaged area is cut apart and repair damaged area, concrete steps are as follows:

A. former breakage image is carried out to pre-service

A.1 image is carried out to Gaussian smoothing

Gaussian smoothing template mask image with following 5 * 5 or 3 * 3:

$\begin{array}{cc}\frac{1}{273}\&times;\left[\begin{array}{ccccc}1& 4& 7& 4& 1\\ 4& 16& 26& 16& 4\\ 7& 26& 41& 26& 7\\ 4& 16& 26& 16& 4\\ 1& 4& 7& 4& 1\end{array}\right]& \frac{1}{16}\&times;\left[\begin{array}{ccc}1& 2& 1\\ 2& 4& 2\\ 1& 2& 1\end{array}\right]\end{array}$

A.2 image gray processing is processed

Adopt mean value method by coloured image gray processing, obtain R, the G of each pixel of image, the mean value of tri-component pixel values of B, and generate its gray level image;

B. damaged area is cut apart

The dividing method that adopts region growing, its process is:

B.1. adopt the method for man-machine interaction, in being repaired the damage zone of image, choose arbitrarily a bit as Seed Points coordinate (x _r, y _r), and on gray level image, record its pixel value p _r(x _r, y _r);

B.2. create a two-value template image mask onesize with former figure, by pixel value p in template image mask _r(x _r, y _r) being made as 1, rest of pixels is set to 0;

B.3. on gray level image, pass through algorithm 1, draw the some set that in template image mask, pixel value is 1, with regard to the damaged area of corresponding original image, be namely partitioned into the damaged area of original image; Algorithm 1 comprises step: (1) creates a storehouse;

(2) obtain Seed Points (x _r, y _r) and pixel value p _r(x _r, y _r);

(3) with Seed Points (x _r, y _r) centered by, calculate its pixel value p _r(x _r, y _r) and eight neighborhood territory pixel value p _i(x _i, y _i) poor, if | p _r(x _r, y _r)-p _i(x _i, y _i) | < M, (i=1,2,3,4,5,6,7,8), the value of M is determined according to experiment, generally get 10, and on mask opposite position, pixel value is at 0 o'clock, by point (x _i, y _i) being pressed into storehouse, the pixel value on mask relevant position is made as 1;

(4) judge that whether storehouse is empty, if not for sky, take out a pixel from storehouse, is used as it as point (x _r, y _r), go back to step (3), otherwise turn to step (5);

(5) finish;

C. repair damaged area

C.1 obtain the frontier point of damaged area

C.1.1 create a queue, be used for storing damaged area frontier point;

C.1.2 obtain damaged area frontier point: to two-value template image mask, the Wei Yuantu damaged area, region that wherein pixel value is 1, traversal view picture mask image, if the pixel value of pixel is 1, and its eight adjacent pixel values has at least one to be 0, the pixel that is just 1 by pixel value deposits in queue, and the queue finally obtaining is exactly the frontier point of damaged area;

C.2 obtain the point that on border, damaged area, right of priority is the highest and the damaged piece of repairing at first

Ι represents full figure, and Ω represents damaged area, and δ Ω represents the border of damaged area Ω, the unbroken part of Φ presentation video, and p represents the point on the δ Ω of border, damaged area, n _pa unit normal vector at p place, represent the isophot curve direction that p is ordered, Ψ _pbe on border centered by the p point, comprised the damaged piece of putting in point in Ω and Φ; The point that on border, damaged area, right of priority is the highest and the damaged piece computation process of the reparation at first centered by this point are as follows:

C.2.1 on computation bound, put the right of priority of p, by formula (1), calculated,

P(p)＝C(p)D(p) (1)

In formula (1): C (p) represents the confidence factor that p is ordered, D (p) represents the data factor that p is ordered, and C (p) is calculated by formula (2),

$C\left(p\right)=\frac{{\mathrm{\&Sigma;}}_{q\&Element;{\mathrm{\&Psi;}}_p\&cap;(I-\mathrm{\&Omega;})}C\left(q\right)}{\left|{\mathrm{\&Psi;}}_p\right|}---\left(2\right)$

In formula (2): | Ψ _p| be Ψ _parea, in image confidence factor a little according to formula (3) initialization,

$C\left(k\right)=\left\{\begin{array}{cc}0& \&ForAll;k\&Element;\mathrm{\&Omega;}\\ 1& \&ForAll;k\&Element;I-\mathrm{\&Omega;}\end{array}\right.---\left(3\right)$

In formula (1), D (p) through type (4) calculates,

$D\left(p\right)=\frac{|{\&dtri;I}_p^{\&perp;}\&CenterDot;n_p|}{\mathrm{\&alpha;}}---\left(4\right)$

Wherein α is normalized factor, is the maximal value of image gray levels;

C.2.2 on computation bound right of priority a little, thereby and the size of fiducial value obtain the highest point of right of priority, be designated as p ₀;

C.2.3 obtain with p ₀centered by the piece of repairing at first this piece size according to damaged area texture structure around, select and sizable of texture structure, adopt the mode of man-machine interaction, the length of side of piece is the odd number between 3 to 99, i.e. the piece of 3 * 3 to 99 * 99 sizes; Then enter step c 3; If image have left and right, upper and lower, upper right with all directions of lower-left, upper left and bottom right to one of symmetrical situation, select based on all directions to balanced sample piece restorative procedure, otherwise be chosen in the restorative procedure that any direction is searched for balanced sample piece;

C.3 search in any direction best balanced sample piece or in the image repair of the best balanced sample pieces of eight direction findings

C.3.1 find in any direction best balanced sample piece

C.3.1.1 in I-Ω with arbitrfary point p ₁centered by sample block choose

P ₀the central point of repairing at first piece, p ₁be the central point of sample block to be found, obtain p ₁and p ₀line and horizontal direction angle theta, utilize formula (5) to obtain p ₁rotate to and p ₀p on same level line position ₂coordinate:

$\left\{\begin{array}{c}{x}_2=(x_1-x_0)\cos \mathrm{\&theta;}+(y_1-y_0)\sin \mathrm{\&theta;}+x_0\\ y_2=-(x_1-x_0)\sin \mathrm{\&theta;}+(y_1-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(5\right)$

In formula (5): x ₀, x ₁and x ₂respectively p ₀, p ₁and p ₂horizontal ordinate, y ₀, y ₁and y ₂respectively p ₀, p ₁and p ₂ordinate;

Acquisition is with p ₂centered by, size and identical and with become the sample block of horizontal symmetrical take 3 * 3 as example, the corresponding relation of damaged piece and sample block is as shown in Fig. 4 and Fig. 5, and Fig. 4 is damaged piece fig. 5 represents horizontal symmetrical sample block two piece symmetric positions of numeral wherein; Recycling formula (6) obtains sample block the position at opposite spin θ angle:

$\left\{\begin{array}{c}x\text{'}=(x-x_0)\cos \mathrm{\&theta;}-(y-y_0)\sin \mathrm{\&theta;}+x_0\\ y\text{'}=(x-x_0)\sin \mathrm{\&theta;}+(y-y_0)\cos \mathrm{\&theta;}+y_0\end{array}\right.---\left(6\right)$

In formula (6): before (x, y) represents reverse rotation the coordinate of middle arbitrfary point, (x ', y ') represent behind rotation θ angle the coordinate of middle corresponding point; after rotation, with point corresponding relation and with corresponding relation identical, obtain with p ₁centered by balanced sample piece;

C.3.1.2 at I-Ω, search for best balanced sample piece, by formula (7), calculated,

$\mathrm{\&Psi;}\hat{p}=\arg \underset{p_1\&Element;I-\mathrm{\&Omega;}}{\min }d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)---\left(7\right)$

In formula (7): represent best balanced sample piece, with p ₁centered by sample block, represent multiblock to be repaired at first with balanced sample piece between similarity measurement, by formula (8), calculate,

$d(\mathrm{\&Psi;}{p}_0,\mathrm{\&Psi;}{p}_1)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2}---\left(8\right)$

In formula (8): x _ijrepresent the pixel value of mid point, y _ijrepresent the pixel value of mid point, the size of piece determines the value of m, if select the size of piece, is 3 * 3, the value of m is 3;

C.3.1.3 the renewal of damaged piece

C.3.1.3.1 according to the corresponding relation in Fig. 4 and Fig. 5, by damaged piece in best balanced sample piece for damaged pixel value in respective pixel value replace;

C.3.1.3.2 upgrade the confidence factor of previous step filler pixels point, by formula (9), calculated,

$\begin{array}{cc}C\left(p\right)=C\left(\hat{p}\right)& \&ForAll;p\&Element;\mathrm{\&Psi;}\hat{p}\&cap;\mathrm{\&Omega;}\end{array}---\left(9\right)$

The pixel value of corresponding point in mask on former breakage image filling point position is updated to 0 simultaneously;

C.3.1.4 upgrade the area of damaged area, calculate the pixel count in damaged area after filling

Every filling once will be recalculated the area of damaged area afterwards, and statistics is filled the number of pixels of rear damaged area, to judge whether repair process finishes; If damaged area area is zero, reparation completes, otherwise repeats to do step c 1, step c 2 and step c 3.1;

C.3.2 at the best balanced sample piece of eight direction findings

Eight arrows represent the damaged piece Ψ centered by p _peight directions, from these eight direction finding balanced sample pieces, each direction is found out a similar sample block of symmetry, then from wherein finding out best balanced sample piece, Ψ _qΨ _pbest balanced sample piece; Calculation procedure is as follows:

C.3.2.1 at eight similar balanced sample pieces of direction finding

From eight directions, find symmetrical similar sample block by formula (10), calculated:

$\begin{array}{cc}\mathrm{\&Psi;}{\hat{q}}_i=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(10\right)$

In formula (10): represent Ψ _pwith similarity measure, subscript 1,2,3,4,5,6,7 and 8 represent respectively a left side, upper left, upper, upper right, the right side, bottom right, under, lower left; 1, the Comparability of 2,3 and 4 four directions formula (11) calculating for amount, the Comparability of 5,6,7 and 8 four directions is formula (12) calculating for amount:

$\left\{\begin{array}{c}{d}_1(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_1=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_2(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_2)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_3(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_3)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_4(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_4)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(11\right)$

$\left\{\begin{array}{c}{d}_5(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_5=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{i(m-j+1)})}^2})\\ d_6(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_6)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-j+1)(m-i+1)})}^2}\\ d_7(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_7)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{(m-i+1)j})}^2}\\ d_8(\mathrm{\&Psi;p},\mathrm{\&Psi;}{q}_8)=\sqrt{\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ij}}-y_{\mathrm{ij}})}^2}\end{array}\right.---\left(12\right)$

In formula (11), (12): x _ijrepresent Ψ _pthe pixel value of mid point, y _ijrepresent the pixel value of mid point;

C.3.2.2 the calculating of best balanced sample piece

From eight similar sample block of symmetry, find Best similarity balanced sample piece, according to formula (13), calculate:

$\begin{array}{cc}\mathrm{\&Psi;q}=\arg \min d_i(\mathrm{\&Psi;p},\mathrm{\&Psi;}{\hat{q}}_i)& (i=\mathrm{1,2,3,4,5,6,7,8})\end{array}---\left(13\right)$

C.3.2.3 upgrade the damaged pixel of damaged piece

C.3.2.3.1 according to following rule Ψ _qin pixel value upgrade Ψ _pmiddle damaged pixel, when pixel is at Ψ _pin scanning sequency be from left to right, from top to bottom time, corresponding Ψ _qin scanning sequency be: when the direction at best balanced sample piece place is 1 or 5, pixel scanning sequency is from right to left, from top to bottom in best balanced sample piece; When the direction at best balanced sample piece place is 2 or 6, in best balanced sample piece, pixel scanning sequency is from top to bottom, from right to left; When the direction at best balanced sample piece place is 3 or 7, in best balanced sample piece, pixel scanning sequency is from left to right, from top to bottom; When the direction at best balanced sample piece place is 4 or 8, in best balanced sample piece, pixel scanning sequency is from top to bottom, from left to right; And by Ψ _pmiddle damaged pixel Ψ _qthe pixel value of middle correspondence is filled;

Take block size 3 * 3 as example, optimal sample piece left to time, Fig. 7 has described the process that damaged piece pixel is upgraded, wherein a represents damaged piece Ψ _p, b represents best balanced sample piece Ψ _q, the numeral scanning sequency in a, b, the corresponding pixel of digital identical expression, if Ψ _pin 4 belong to damaged pixel, just use Ψ _qthe pixel value of middle position 4 is filled;

C.3.2.3.2 upgrade the confidence factor of previous step filler pixels point, by formula (9), calculated; The pixel value of corresponding point in mask on image filling point position is updated to 0 simultaneously;

C.3.2.4 upgrade the area of damaged area, if damaged area area is zero, reparation completes, otherwise repeats to do step c 1, step c 2 and step c 3.2.

2. a kind of image repair method based on balanced sample piece as claimed in claim 1, is characterized in that in step c 2.3, selects 9 * 9,11 * 11 to 33 * 33 piece.